Ways to automate engineering work

Compare how local coding agents, agent automations, CI platforms, and Shipfox run and coordinate engineering work.

Swipe to compare.

Compare approaches to engineering work
Comparison point	Local coding agent	Agent automations	CI platform	Software factory (Shipfox)
Example providers
Products	Claude Code Codex CLI OpenCode	GitHub Agentic Workflows Claude Code Routines Cursor Automations	GitHub Actions GitLab CI/CD	Shipfox
Running work
Where code runs	Engineer's machine	Provider-managed cloud runner	Hosted or self-hosted runners	Hosted or self-managed runners
Execution environment	No runner choice	Preset runner environment	Custom runners, images, and network	Cloud runners in a range of sizes Self-hosted runners on your infrastructure
Parallel work	No managed parallel jobs	Parallel managed runs	Parallel jobs within runner capacity	Parallel jobs by default
Triggers	Human starts a session	Provider-supported events and schedules	Repository events, schedules, and webhooks	8 built-in event sources, 132 event types Cron schedules Manual triggers
Coordinating agents and steps
Later review comments	Engineer follows up in chat	New automation run	New workflow run	Listening job continues the same run
Failed check needs a code fix	Engineer prompts another attempt	No independent feedback gate	Checks fail jobs; no built-in agent repair	Gate restarts agent with check feedback
Agent and fixed steps in one flow	No workflow orchestration	No built-in mixed-step workflow	Agent call needs job scripting	Agent, run, and tool steps
Agent conversation across steps	No shared workflow session	No cross-run agent session	No native agent session across jobs	Named session across steps and jobs
Team control
Shared team process	No shared run process	Reusable saved automations	Shared jobs and workflows	Shared workflow definitions
Configuration in code	No versioned workflow definition	Usually configured in provider UI	Workflow files in the repo	Workflow YAML in the repo
Model choice	Agent's supported models	Provider-supported models	Chosen in the agent call	Broad selection of managed models
Agent-ready external tools	Engineer adds MCP servers	Small native action set; MCP for more	Agent tools need custom wiring	8 tool integrations, 134 tools
Tool credentials and scope	Personal credentials	MCP often uses personal OAuth and grants all server tools	Secrets passed to jobs	Workspace credentials; selected tools per step
Visibility and cost
Activity logs	No shared run history	Automation session logs	Workflow and job logs	Run, step, and agent logs
Model cost by task	No shared cost per task	Usage in provider account	Model cost separate from runner usage	Agent usage and cost per step

Was this page helpful?

Edit this page on GitHub